A lithographic photomask, or reticle, is conventionally used for patterning wafers. Deviations of feature sizes or CDs (critical dimensions) on the wafer from their respective target values can limit the process window and the yield, thereby increasing cost, particularly for double patterning applications, which have very small error margins for CDs. By optimizing the exposure dose, an overall best match between the printed CDs and the target CDs may be achieved. However, if the deviations differ substantially for various features within the pattern, the CD deviations of the different features may still limit process window and yield after such an overall dose optimization. Such a situation may be caused by multiple factors. For example, CD variations may occur across the reticle and then be transferred to the wafer during exposure with the reticle. Also, there may be pattern density variations or mistakes in the design or there may be non-optimum data sizing. For these reasons, for designs with different structures, such as multipurpose wafers as used for 28 nanometer (nm) IP shuttles (i.e., devices containing data from different customers for testing and qualification purposes) and devices containing both memory (e.g., SRAM) and logic, it may be difficult to pattern all of the different structures with correct target CDs.
Another problem is that test structures with small sizes, which are printed on the wafer, may collapse or be lifted during the subsequent wafer processing, thereby causing defects on the wafer. In this case, it is desirable to eliminate these structures on the wafer, in other words, to reduce their CD to zero or, alternatively, to increase their CD substantially to make them more stable. Finally, in product development it can be useful to selectively vary the sizes of some specific structures on the wafer to test the impact of such variations, for example, on performance or yield.
CD variations may be corrected by replacing the reticle with a corrected one. However, ordering a new reticle is expensive and adds significant time to the setup time, as no wafers may be printed while the new reticle is being produced. Further, a new reticle must be qualified again (i.e., checked again for errors) thereby adding more time to the setup time. Alternatively, CD variations may be corrected by modifying the existing reticle. However, such reticle correction requires an expensive tool and running costs, and no wafers may be patterned while the reticle is being corrected. Further, with the currently available ways to modify the CD of an already existing reticle, the reticle may be damaged, or registration may fail, new errors may be introduced by the modification, and all changes are usually irreversible. Long range CD variations can be corrected by the exposure tool (with the so-called “dose mapper” function), but both the spatial resolution and the number of degrees of freedom of this correction method are limited.
A need therefore exists for methodology enabling correction of CD variations which are reproducible and which similarly occur in every image field on every wafer, and enabling deliberate variations of CD, without replacing or modifying the reticle.